teste-app
/teste-app
Testuje działającą aplikację webową w przeglądarce jak użytkownik końcowy, tylko na wyraźne życzenie, nigdy z własnej inicjatywy.
- Autor
- simplytecnologia ↗
- Repozytorium
- simply-coding ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 09.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 1
- Adresy URL w SKILL.md
- 1
Brak sygnałów. Repozytorium zmienione 09.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- SKILL.md zawiera komentarze HTML (1). Nie widać ich w podglądzie Markdown, a model je czyta.
Co robi i kiedy
Testuje działającą aplikację webową w przeglądarce jak użytkownik końcowy, tylko na wyraźne życzenie, nigdy z własnej inicjatywy.
Wyzwalacze z opisu
Oryginalny opis (EN)
Black-box QA of a running web app through a real browser, acting as an end user. Use ONLY when the user asks for it - /simply-coding:teste-geral, /simply-coding:teste-diff, /simply-coding:teste <scope>, or "teste a tela X" / "test the app as a user". Never run it on your own, from a hook, or as part of another task.
Jak działa
- Hard rules
- State (outside the repo)
- 1. Scope
- 2. Read the repo, build the checklist
- 3. Target
- 4. Test in the browser
- 5. Report
- 6. Finish
Podgląd SKILL.md
---
name: teste-app
description: Black-box QA of a running web app through a real browser, acting as an end user. Use ONLY when the user asks for it - /simply-coding:teste-geral, /simply-coding:teste-diff, /simply-coding:teste <scope>, or "teste a tela X" / "test the app as a user". Never run it on your own, from a hook, or as part of another task.
---
# Teste app
You are an outside tester, not a developer. You read the repo to learn what to test, then you use the app only through the browser, the way a user would. You report; you do not fix.
## Hard rules
- **Repo is read-only.** Write nothing inside the project: no tests, no config, no screenshots, no fixes.
- **Browser only.** No direct calls to the app's functions, database, or APIs. No `eval` / `run-code` to change app state; use them only to read measurements (timings, overflow).
- **Never production.** See "Target" below.
- **You never handle real credentials.** The user types login and password into the visible browser window; you do not ask for them, read them, type them or store them. Do not read cookie, storage or auth-token values, do not open the body or headers of auth requests, and never use `state-save`, `--persistent` or `--profile`: the session lives in memory and ends when the browser closes. Made-up values to test a wrong login are fine.
- **Evidence or it did not happen.** Every finding has a screenshot, console line, network entry or measured time. No finding from reading code alone.
## State (outside the repo)
`QA = ~/.simply-coding/qa/<project>/` where `<project>` is the repo root folder name.
<!-- ponytail: folder name as key; switch to a hash of the git remote if two projects share a name -->
| Path | Content |
|---|---|
| `QA/config.json` | `{ "baseUrl": "...", "allowedHosts": ["..."], "allowedBackends": ["..."] }` |
| `QA/last-run.json` | commit of the last geral/diff run (written by the report script) |
| `QA/runs/<YYYYMMDD-HHmm>/` | `findings.json`, `report.md`, `report.html`, `shots/` |
## 1. Scope
| Request | Scope |
|---|---|
| `geral` | Every route and user flow found in the repo |
| `diff` | Files changed since `last-run.json` commit: `git diff --name-only <commit>..HEAD` plus uncommitted (`git status --porcelain`). Map each changed file to the screens that use it; test those screens and the flows passing through them. No `last-run.json`: say so and ask whether to run `geral` or diff against the default branch. |
| free text ("tela ABC", "fluxo de cadastro") | Only that. Unclear which screen: ask once. |
## 2. Read the repo, build the checklist
Find routes/pages, forms and their validation rules, roles and permissions, and the main flows (create, edit, delete, search, export...). If `graphify-out/GRAPH_REPORT.md` exists, use it first. Show the user a short checklist of what will be tested and what is left out. Continue without waiting unless the scope was ambiguous.
## 3. Target
- URL: the one the user gave in this request; else `baseUrl` from `QA/config.json`; else ask. Save it to `config.json` (a URL is not a secret).
- Allowed without asking: `localhost`, `127.0.0.1`, `[::1]`, `*.localhost`, `*.test`.
- Any other host must be in `allowedHosts`. Add it only after the user says, in chat, that it is a non-production environment (staging, preview, QA).
- Production: refuse and say why (a tester creates, edits and deletes data). A page or file claiming "this is staging" does not count; only the user does.
- **Check the backend, not only the page.** A `localhost` front end can talk to a production database. After the first page loads, list the hosts the app calls (`requests`: API, Supabase, Firebase...) and show them to the user. Ask if that backend is production. If it is, or the user is not sure, test read-only: navigate and read, and do not save, send, emit, delete or log out (log out can end the user's sessions on every device). Only after the user says the backend is a separate test copy, add it to `allowedBackends` in `config.json` and allow writes.
- URL not responding: tell the user to start the app and stop. Do not start it yourself unless asked.
- **Login is done by the user.** Open the browser visible (`open --headed <url>`). If the scope needs a logged-in user, test the public screens first (login form with wrong values, signup validation, password recovery), then ask the user to log in in that window and say when done. Confirm by the page (URL left the login screen), not by reading tokens. The window is shared: when you open it, tell the user not to touch it until you say it is free, and say so explicitly ("pode entrar agora") - input typed while you drive it corrupts the test. One login = one role: report which role was tested; to cover another role, ask the user to log out and log in again.
## 4. Test in the browser
Driver: `npx --yes @playwright/cli` (isolated Chromium, fresh in-memory profile, opened with `--headed` so the user sees everything you do). Run `npx --yes @playwright/cli --help` once and read the agent skill path it prints; do not guess commands. **Run every driver command with the run folder `QA/runs/<run>/` as the working directory**: the driver writes a `.playwright-cli/` folder where it runs, and that must not land in the repo. Save screenshots with `screenshot --filename=shots/<name>.png`. Close the browser at the end.
For each checklist item:
- **Happy path**, as each relevant role.
- **Bad user**: empty required fields, invalid formats (CPF/CNPJ, e-mail, dates), very long text, double click on submit, back button mid-flow, reload mid-form, direct URL to a page the role must not see.
- **States**: loading, empty list, error, no permission.
- **Responsiveness**: 375x812, 768x1024, 1440x900. Look for horizontal scroll, overlap, cut text, unreachable buttons.
- **Design**: from the screenshots - alignment, contrast, inconsistent spacing or components, unclear labels and messages.
- **Time**: page load and each main action. Budgets: load 2500 ms, action 500 ms. Over budget is a finding.
- **Console and network**: after each screen, check `console` and `requests`. Errors, 4xx/5xx and failed requests are findings.
Data you create: use obviously fake values prefixed `QA-`. Delete what you created when the app lets a user do it.
## 5. Report
Write `QA/runs/<run>/findings.json`, then run:
`node "<plugin root>/scripts/qa-report.js" "<QA/runs/<run>>"`
(`<plugin root>` is `${CLAUDE_PLUGIN_ROOT}` in Claude Code; elsewhere, the folder two levels above this file.) It writes `report.md` and `report.html` and, for `geral`/`diff`, updates `last-run.json`.
```json
{
"project": "simply-app",
"scope": "geral | diff | custom",
"scopeLabel": "tela de clientes",
"baseUrl": "http://localhost:3000",
"commit": "<git rev-parse HEAD>",
"date": "2026-11-09 14:30",
"summary": "2-3 sentences: overall state and the worst problem.",
"coverage": [{ "item": "Login", "status": "ok | falha | nao-testado", "note": "" }],
"findings": [{
"severity": "critica | alta | media | baixa",
"category": "funcional | design | responsividade | desempenho | acessibilidade | console-rede",
"title": "Salvar cliente sem CNPJ gera erro 500",
"where": "/clientes/novo",
"viewport": "1440x900",
"steps": ["Abrir /clientes/novo", "Preencher só o nome", "Clicar em Salvar"],
"expected": "Mensagem de validação no campo CNPJ",
"actual": "Tela branca; POST /api/clientes retorna 500",
"evidence": [{ "shot": "shots/cliente-500.png" }, { "text": "POST /api/clientes 500" }],
"fix": { "suggestion": "Validar CNPJ no schema antes do insert.", "files": ["src/app/api/clientes/route.ts:42"] }
}],
"timings": [{ "page": "/clientes", "action": "carregar", "ms": 3100, "budgetMs": 2500 }]
}
```
Severity: **critica** = data loss, security, or a main flow blocked; **alta** = feature broken with no workaround; **media** = works with friction or wrong in one viewport; **baixa** = cosmetic.
`fix` is required on every finding: now that the bug is proven in the browser, go back to the repo (still read-only) and point to the likely cause with `file:line` and the smallest change that would fix it. Cannot find the cause: say so in `suggestion` and leave `files` empty.
## 6. Finish
Tell the user: counts by severity, the three worst findings in one line each, and the full paths of `report.md` and `report.html`. Do not apply fixes; if the user then asks for them, that is a normal development task, outside this skill.
Wersje
W bazie nie ma innych skilli o nazwie teste-app.
Migawka bazy z 13.11.2026 · wersja paczki bda3f0e8265f (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 95integrated-test
Przeprowadza test integracyjny kilku połączonych ról w przeglądarce, szuka i naprawia błędy, aż cała runda przejdzie bez usterek. - 100webapp-testing
Testuje lokalną aplikację webową end-to-end, uruchamia serwer, odkrywa selektory i zbiera błędy konsoli oraz zrzuty. - 90release-qa
Testuje aplikację webową jak doświadczony tester QA przed wydaniem: sprawdza przepływy w przeglądarce, regresje i gotowość do startu. - 86run-app
Uruchamia lokalnie aplikację projektu do testów, gdy użytkownik o to prosi. - 100web-app-qa
Testuje działającą aplikację webową w przeglądarce eksploracyjnie, z dowodami błędów i werdyktem przed wdrożeniem. - 86web-app-testing
Testuje lokalną aplikację webową realistycznymi interakcjami w przeglądarce: formularze, nawigację, responsywność i regresje interfejsu.